Flagging as an external, unaffiliated contributor — advisory only, no write access to this repo. Verified locally against a fresh clone; no code changes proposed here per the new contribution policy in discussion #405.
assets/testnet/config.json and assets/devnet/config.json no longer satisfy schemaGenesisConfig, so genesis regeneration for both networks throws instead of producing output. assets/mainnet/config.json is unaffected.
Repro (no RPC/mnemonic needed — the parse happens before either is used):
git clone --recurse-submodules https://github.com/circlefin/arc-node
cd arc-node && npm ci
foundryup -i $(cat .foundry-version) && npx hardhat compile
npx hardhat genesis --network testnet # or --network devnet
Both throw a ZodError with the same six issues:
ProtocolConfig: Unrecognized key(s) in object: 'beneficiary'
ValidatorManager.validators.0.controllers: Required
ValidatorManager.validators.1.controllers: Required
ValidatorManager.validators.2.controllers: Required
ValidatorManager.PermissionedValidatorManager.pauser: Required
ValidatorManager.PermissionedValidatorManager: Unrecognized key(s) in object: 'controllers'
(Separately, the Hardhat CLI's own error handler crashes with an unrelated TypeError: Cannot read properties of undefined (reading 'value') while trying to print this ZodError, from node:internal/util/inspect — that's a cosmetic issue in the CLI's error formatting, not part of this bug. Calling the network's genesis.config.ts build() directly surfaces the real ZodError above.)
Root cause: both committed config files predate two schema changes that have since landed in the public repo:
46324c4 (2026-05-05) dropped ProtocolConfig.beneficiary.205c9f2 (2026-05-21) restructured ValidatorManager: PermissionedValidatorManager.controllers: address[] became a required pauser: address field, and each entry in validators[] gained a required controllers: {address, votingPowerLimit}[] array (replacing the flat top-level controller list).
assets/testnet/config.json and assets/devnet/config.json were last written by d3ed5c7 (2026-04-22) — before either change — and have not been regenerated since. assets/mainnet/config.json was authored later in the new shape, which is why only mainnet still parses. The two most recent schema commits (#346, #359, both 2026-09-08) are unrelated to this — they only add stricter numeric bounds that testnet/devnet's existing values already satisfy.
Impact: CI doesn't invoke genesis for testnet/devnet (only localdev and mainnet are exercised anywhere in .github/workflows/), so this is silent rather than a build break. It surfaces whenever someone tries to actually regenerate testnet/devnet genesis locally, or treats these committed configs as a source of truth for the current validator/pauser/controller wiring on those networks — the "controllers" they list are stale under the current schema's semantics.
On a fix: this isn't a schema bug to patch — the config files need to be brought up to the new shape with real values (which pauser address, which per-validator controller(s) at which voting-power limits) that only the team can supply for testnet/devnet, the way assets/mainnet/genesis.config.ts already hardcodes real addresses for mainnet. Happy to take this if it gets assigned; per the new policy I won't open a PR without that.
Flagging as an external, unaffiliated contributor — advisory only, no write access to this repo. Verified locally against a fresh clone; no code changes proposed here per the new contribution policy in discussion #405.
assets/testnet/config.jsonandassets/devnet/config.jsonno longer satisfyschemaGenesisConfig, so genesis regeneration for both networks throws instead of producing output.assets/mainnet/config.jsonis unaffected.Repro (no RPC/mnemonic needed — the parse happens before either is used):
Both throw a
ZodErrorwith the same six issues:(Separately, the Hardhat CLI's own error handler crashes with an unrelated
TypeError: Cannot read properties of undefined (reading 'value')while trying to print this ZodError, fromnode:internal/util/inspect— that's a cosmetic issue in the CLI's error formatting, not part of this bug. Calling the network'sgenesis.config.tsbuild()directly surfaces the real ZodError above.)Root cause: both committed config files predate two schema changes that have since landed in the public repo:
46324c4(2026-05-05) droppedProtocolConfig.beneficiary.205c9f2(2026-05-21) restructuredValidatorManager:PermissionedValidatorManager.controllers: address[]became a requiredpauser: addressfield, and each entry invalidators[]gained a requiredcontrollers: {address, votingPowerLimit}[]array (replacing the flat top-level controller list).assets/testnet/config.jsonandassets/devnet/config.jsonwere last written byd3ed5c7(2026-04-22) — before either change — and have not been regenerated since.assets/mainnet/config.jsonwas authored later in the new shape, which is why only mainnet still parses. The two most recent schema commits (#346,#359, both 2026-09-08) are unrelated to this — they only add stricter numeric bounds that testnet/devnet's existing values already satisfy.Impact: CI doesn't invoke
genesisfor testnet/devnet (onlylocaldevandmainnetare exercised anywhere in.github/workflows/), so this is silent rather than a build break. It surfaces whenever someone tries to actually regenerate testnet/devnet genesis locally, or treats these committed configs as a source of truth for the current validator/pauser/controller wiring on those networks — the "controllers" they list are stale under the current schema's semantics.On a fix: this isn't a schema bug to patch — the config files need to be brought up to the new shape with real values (which
pauseraddress, which per-validator controller(s) at which voting-power limits) that only the team can supply for testnet/devnet, the wayassets/mainnet/genesis.config.tsalready hardcodes real addresses for mainnet. Happy to take this if it gets assigned; per the new policy I won't open a PR without that.